That's not what the author said. The original scenario that he proposed had data being generated chaoticly. As that data was generated, the author calculated the probability/improbability of 41 characters sequencing in the correct order in any part of that entire data stream. Thus, for the desired result, there is no "historical vaccum" because the entire history of the data is being examined by the math.
As for your comment on "feedback systems", you've got two large hurdles to overcome: First, you have to show that a particular feedback system was in place prior to the very first life form becoming animated, and Second, you've got to demonstrate that such a feedback mechanism could actually change the probability / improbability of useful data self-forming from a chaotic environment.
Frankly, I doubt that either of those two hurdles could be overcome, and even if they were properly managed, would probably not alter the final mathematical result substantially.
Someone also made the flawed comment that the math didn't account for data forming in groups, and then those groups joining with each other, but that too is addressed in the math. To wit: whether you flip four coins at once or one coin at a time, the odds of getting all heads remains precisely the same. So too with data self-forming from a chaotic environment